_{Building tools for AI coding agents.}

Currently exploring multi-agent simulation, long-term memory for LLMs, and the developer infrastructure that makes AI coding assistants observable, memorable, and verifiable.

Neutral orchestration and collaboration layers for AI coding agents.

• mco — Orchestrate AI coding agents. Any prompt × any agent × any IDE — works with Claude Code, Codex CLI, Gemini CLI, OpenCode, and Qwen Code, from Cursor, Trae, Copilot, Windsurf, or plain shell. Python
• squad — Multi-AI agent terminal collaboration tool. Rust

A curated collection of skills I've authored for Claude Code.

• code-review-skill — Comprehensive code review skill covering React 19, Vue 3, Rust, TypeScript, TanStack Query v5, and more.
• first-principles-skill — Systematic first-principles thinking — analyze designs, challenge assumptions, build from fundamental truths.
• 5-whys-skill — Five-Whys root cause analysis based on Toyota Production System methodology.
• mobile-app-design — Mobile UI/UX design skill — iOS, Android, accessibility, React Native best practices.
• mermaid-syntax-skill — Generate error-free Mermaid diagrams; handles special characters, reserved words, and escaping.
• memex — Local full-text search engine for Claude Code and Codex session history.
• manim-skill — Manim animation skill — create programmatic animations for technical blogs and educational content.

Make AI agents more observable, memorable, and verifiable.

• agmon — htop for AI Agents. Real-time token cost and tool-call observability across Claude Code and Codex, with TUI and web dashboard. Go _{Single Go binary; auto-installs Claude Code hooks on first run; ships weekly Markdown cost reports.}
• archify — Claude Skill that turns plain-language descriptions into export-ready architecture diagrams with dark/light themes and 4× PNG/SVG export.
• agent-inspect — Parallel multi-agent code audit slash command for Claude Code, OpenCode, and Codex CLI.
• evermemos-mcp — MCP memory server giving AI coding assistants persistent cross-session recall — benchmarked 60/60 accuracy on EverMemOS Cloud. Python · MCP
• vscode-ai-commit — VS Code extension that streams AI-generated commit messages from staged diffs with commitlint awareness. TypeScript

From multi-agent worlds to AI-powered editors and durable memory.

• MiroFish-local — Local-privacy edition of the MiroFish multi-agent social simulation. Replaces Zep Cloud with Graphiti + Neo4j; runs fully air-gapped. Python · Neo4j _{Single env var toggles between Zep Cloud and air-gapped Graphiti + Neo4j memory.}
• chat_edit — Dual-mode AI app combining streaming chat and rich-text AI editing, with Mermaid, Monaco diff, multi-format export, offline-first storage. Vue 3.5 · TypeScript
• mnemo — AI assistant with three-layer durable memory (episodic / semantic / procedural) that reconstructs full session context from EverMemOS Cloud. Next.js · FastAPI

Built to ship — or to understand.

• coding-cli-guide — 158-page interactive visual guide to Gemini CLI internals: Agent Loop, tool scheduling, security sandboxing, MCP. Live demo. React 19 · Vite 7 _{120+ step-by-step animations across 13 topic groups; deep-linkable per topic.}
• doppel — macOS CLI/TUI that clones any .app bundle into a runnable second instance with a new bundle ID and ad-hoc re-signing. Useful for running multiple copies of Electron-based AI tools side by side. Go _{Single binary; TUI for interactive use, CLI with --json for scripting; brew install tt-a1i/tap/doppel.}
• pkg-analyzer — Published npm CLI that visualizes node_modules size, duplicates, and unused deps, with clipboard export for AI-powered optimization. TypeScript

PRs to projects I use and depend on.

• JS tooling & runtimes — biome, oxc, bun, deno, TypeScript, rspack
• Web frameworks & libraries — fastify, elysia, TanStack/form, tRPC, drizzle-orm
• AI coding ecosystem — gemini-cli, qwen-code, agentgateway, kgateway
• Apps & infrastructure — Apache Superset, react-grab